AI-Powered Career Intelligence & Learning Ecosystem for Healthcare Technology
SkillOrbit is a comprehensive, AI-driven career guidance platform that leverages a hybrid AI architecture to help users discover personalized career paths, identify skill gaps, and navigate their professional journey through intelligent recommendations and structured learning roadmaps.
In today's rapidly evolving job market, identifying the right career path is challenging. SkillOrbit addresses this by analyzing user intent beyond simple keywords, providing a data-driven "GPS" for professional growth.
- Intelligent Interest Detector: A sophisticated 7-factor assessment analyzing cognitive styles, technical depth, work environment, and impact motivation.
- Semantic Career Matching: Harnesses BERT embeddings to discover ideal roles based on the "meaning" of your background, not just keyword matches.
- Dynamic AI Roadmaps: Generates on-demand learning paths with modular milestones, sub-topics, and estimated durations using Mistral AI.
- Hyper-Sensitive Skill Gap Analysis: Uses micro-contextual chunking to identify exactly which skills you possess and which ones are target gaps.
- Integrated Course Recommendations: Automatically matches learning modules with specialized Coursera courses using vector search (Supabase pgvector).
- Immersive 3D Experience: Cinematic hero section powered by Spline, creating a futuristic and engaging user entry point.
- Interactive Dashboard: Manage up to 3 active roadmaps simultaneously with real-time progress tracking and persistent state.
SkillOrbit uses a specialized two-tier AI system to balance logical reasoning with precise semantic understanding.
- Model:
Xenova/all-MiniLM-L6-v2(Quantized q8) - Library: Hugging Face Transformers.js
- Execution: On-Demand Local Inference via
@huggingface/transformers. - Purpose: Vectorization and Semantic Search.
- How it works: It converts user profiles and course descriptions into 384-dimensional vectors. Unlike keyword search, BERT understands that "Distributed Systems" is semantically related to "Scalability," even if the words don't match.
- Difference with Mistral: BERT doesn't "think" or "chat"; it maps text into a high-dimensional mathematical space for lightning-fast similarity comparisons in Supabase.
- Model:
open-mistral-7b - Execution: Cloud API.
- Purpose: Complex Reasoning and Content Generation.
- How it works: It takes the "context" (user answers + top semantic matches) and reasons through them to build a logical 5-7 module roadmap. It handles the "Why" (reasoning) and the "How" (curriculum structure).
The core engine of SkillOrbit is its Hyper-Sensitive Skill Recognition algorithm. Unlike traditional platforms that use basic keyword matching, SkillOrbit treats your profile as a set of multidimensional semantic features.
To ensure no skill is missed, the algorithm breaks down the user's profile into granular "Micro-Contexts":
- Skills & Education: Direct mapping of known skills and academic background.
- Project Semantic Extraction: Project descriptions are split into individual sentences. This ensures a mention of "leveraged Git for version control" in a 300-word paragraph is isolated as a specific evidence point.
Each job requirement is analyzed against the user's "Knowledge Base" using a two-tier approach:
- Tier 1: Literal Match: Fast-path substring comparison for tool names (e.g., "React" matching "React.js").
- Tier 2: Semantic Similarity (BERT):
- The requirement (e.g., "Cloud Infrastructure") and the user's context (e.g., "Deployed to AWS") are vectorized using all-MiniLM-L6-v2.
- Cosine Similarity is calculated. A score > 0.60 counts as a recognized skill, even if the terminology differs.
Unmatched skills are categorized into a hierarchical gap report:
- Foundational Gaps: Missing core skills required to start the career.
- Intermediate/Advanced Gaps: Skills needed for professional growth.
- Result: This data is passed to Mistral AI to ensure the generated Roadmap prioritizes foundational gaps first.
SkillOrbit doesn't just tell you what's missing; it provides the bridge to acquire those skills through a high-performance vector search system.
- Source: A curated and scraped dataset of specialized courses from Coursera.
- Attributes: Includes course titles, providers (e.g., Google, IBM, Stanford), and detailed descriptions.
- Vectorization: The entire corpus is pre-processed and stored as BERT embeddings within the database.
- Technology: pgvector extension on Supabase.
- The Advantage: Instead of standard SQL
LIKEqueries which fail with context, we use Neighbor Search (Vector Similarity). - Indexing: Uses an HNSW (Hierarchical Navigable Small World) index for sub-millisecond similarity search across the entire course library.
When a learning module is generated:
- Context Synthesis: A weighted search query is built using the
Module Title+Missing Skills. - On-Demand Embedding: The BERT model (
all-MiniLM-L6-v2) creates a 384D vector representing the intent of the module. - Cross-Reference: The algorithm performs an RPC call (
match_coursera_courses) in Supabase to calculate the distance between the module's intent and the course library. - Curated Suggested: Only courses with a similarity score > 0.35 are recommended, ensuring high relevance to the specific career gap.
#Extra Features
- Next.js 15: App Router architecture for optimized performance and SEO.
- TypeScript: Full-stack type safety.
- Tailwind CSS: Modern, utility-first styling with a premium "Glassmorphism" aesthetic.
- Framer Motion: Smooth, cinematic animations and transitions.
- Lucide React: High-quality vector iconography.
- MongoDB (Mongoose): Primary store for user accounts, dashboard state, and progress.
- Supabase (pgvector): High-performance vector database for semantic course matching.
- JOSE / bcryptjs: Secure JWT-based authentication and password hashing.
- Mistral AI API: Large Language Model for logic and generation.
- Hugging Face Transformers: Infrastructure for running BERT models directly in the Node.js runtime (no Python required).
- Spline: 3D design and interaction for the immersive hero experience.
- Node.js 20.x or higher
- MongoDB Atlas account
- Mistral AI API Key
- Supabase Project
# 1. Clone the repository
git clone https://github.com/CodexKnight-ai/SkillOrbit.git
cd skill-orbit
# 2. Install dependencies
npm install
# 3. Setup environment variables
cp .env.example .env
# Edit .env with your credentials (see below)
# 4. Start development server
npm run devCreate a .env file with following keys:
# Database
MONGODB_URI=mongodb+srv://<user>:<password>@cluster0.mongodb.net/skillorbit
NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_anon_key
SUPABASE_SERVICE_ROLE_KEY=your_service_role_key
# AI Configuration
MISTRAL_API_KEY=your_mistral_key
MISTRAL_MODEL=open-mistral-7b
# Authentication
JWT_SECRET=your_32_character_random_secretgraph TD
A[User Interest Detector] --> B[BERT Vectorization]
B --> C{Supabase Vector Search}
C --> D[Ranked Career Candidates]
D --> I[Skill Gap Analysis BERT + Chunking]
I --> E[Mistral AI Roadmap Generation]
E --> J{Semantic Course Search}
J --> F[Final Personalized Roadmap]
F --> G[User Dashboard]
G --> H[Milestone Tracking]
Developed for the Ingenium 2026 Hackathon - Empowering the next generation of healthcare technologists.